US-China AI Talks Expected Ahead of Xi's September Visit to US, Led by Bessent

Reuters reported on Tuesday (21st) citing five知情人士 that the United States and China are expected to hold artificial intelligence (AI) talks in September to discuss how to reduce risks posed by each other's increasingly powerful frontier AI models. This would mark the first formal governmental dialogue on AI between the two nations since President Trump took office.

According to知情人士, the upcoming dialogue is one of the key outcomes of the May summit between President Trump and Chinese President Xi Jinping, and could take place before Xi's scheduled visit to the US on September 24, although the exact date has not yet been finalized. Preparations remain in early stages, with officials, agenda, and location still under discussion.

As the global AI race accelerates, the US and China are not only competing in economic, political, and military domains but are also increasingly concerned about each other's AI capabilities. Both nations are assessing how to regulate advanced AI models to prevent their use in enhancing military power, launching cyberattacks on critical infrastructure, or causing massive disruptions to labor markets.

知情人士 indicated that the US delegation will be led by Treasury Secretary Scott Bessent. The White House, State Department, and Treasury did not immediately respond to requests for comment; neither did China's National Development and Reform Commission, Cyberspace Administration, Ministry of Industry and Information Technology, Foreign Ministry, or State Council.

First official AI dialogue under Trump administration

The US and China have previously engaged on AI safety issues. In 2024, then-President Biden and Xi Jinping agreed after a working-level meeting in Geneva that decisions on nuclear weapon use should be made by humans, not AI. However, if the September talks proceed, it would be the first official US-China AI dialogue during Trump's current term.

The US currently restricts exports of some of the most advanced AI chips to China. Axios previously reported that Washington is considering limiting US companies' use of open-weight AI models from China, amid rapidly growing capability and accessibility of such models.

On the Chinese side, authorities are considering restricting overseas use of its most powerful AI models developed domestically, due to concerns they could be used for hacking attacks. The report specifically mentioned Anthropic's Mythos model, which has not been publicly released, whose potential cyberattack capabilities have raised concerns in Beijing.

Bessent stated in May that future talks would aim to prevent the proliferation of powerful AI models to 'non-state actors,' and suggested such meetings could begin within 4 to 8 weeks. After Trump's visit to China, China's Foreign Ministry confirmed the two countries had agreed to establish a governmental AI dialogue mechanism, though Beijing has not publicly updated progress since.

Beijing hopes to focus on technical issues and avoid politicizing the dialogue

Analysts noted that Beijing's discussion priorities include concerns about Mythos, how the US will regulate future advanced models, and whether Washington will restrict Chinese open-weight models. The Trump administration is currently consulting with US AI companies on developing voluntary safety standards for new model releases, though no specific regulatory measures have been announced.

One知情人士 said Chinese regulatory agencies will continue deliberating AI issues over the next month before deciding which ministry officials will participate in the dialogue. Beijing hopes for substantive technical engagement from the US, rather than allowing talks to become politicized.

Paul Triolo, partner at DGA-Albright Stonebridge Group, believes the two sides are unlikely to resolve major issues in the first meeting. A more realistic outcome would be reaching consensus on basic concepts such as defining 'frontier AI models,' thereby establishing a common language for future risk management and regulatory cooperation.